By François Pachet and Pierre Roy
François Pachet and Pierre Roy. “Hit Song Science Is Not Yet a Science.” Proceedings of ISMIR 2008, Philadelphia, pp. 355–360.
About the paper
Can you predict a hit from the sound of a song or from descriptions of it? This study puts that claim to the test with a catalog of roughly 32,000 tracks.
Abstract in brief
The authors compare two sets of audio features with manually assigned descriptions of tracks. The models learned some subjective descriptors, such as whether a song sounded nostalgic. The features tested did not reliably predict the popularity measures used in the study.
Related research
A 2020 study of streaming skips and musical structure cites this work and looks at when listeners stop playing a track, using observed behavior within each song.